Why AI Will Never Take Over the World AI is currently an evolving technology that is, at best, only partially successful. Nonetheless, people give it more credit than it is due because they tend to fill the void left by their ignorance on the topic with the human propensity to anthropomorphize just about anything and everything on the planet. In brief, AI accepts cleaned data as input, analyzes it, finds the patterns, and provides a requested output. In other words, AI doesn't really "understand" anything. Consequently, it can neither create nor discover anything new. Moreover, it has no intrapersonal knowledge. It simply behaves as designed by human programmers, and what folks think of as intelligence is merely a mix of clever programming and vast amounts of data analyzed in a specific manner. In short, taking over the world is beyond AI's capability, at least for the foreseeable future, simply because Artificial Intelligence cannot suddenly become self-aware. It utterly lacks the means (the requisite categories of intelligences) to do so. And even if it didn't, that still would not be enough, because scientist do not understand (and therefore have no idea how to program) the intangible "spark" that makes people human. And if one accepts the claims made in Dr. Mohler's article (see Post #9), that non-material aspect of humanity is the image of God; and the only One who can program that is God Himself.
Algorithms, Planning and Problem Solving Algorithms An algorithm presents a series of steps, or sequences of operations, designed to find the correct answer to a question in a reasonable amount of time—or report back if no solution is found—but does not necessarily perform all the steps to solve the problem. AI algorithms distinguish themselves from generic algorithms by solving complex problems whose resolution is considered to be typically or exclusively the product of human intellect—problems that are often part of the "NP-complete" (nondeterministic polynomial time) class of problems that humans routinely deal with by using a mix of rationality and intuition. NP-compete problems differ from other algorithmic problems in that finding their solutions in a reasonable time frame is not yet possible, and even if you had computers more powerful that those available today, a search for their solutions would last almost forever. In other words, they are not the kinds of problems one solves by trying all possible combinations or possibilities. Planning The State-Space Search Planning is a classic AI problem that helps you determine the sequence of actions to perform to achieve a certain goal. One way to accomplish this is to start from the present state, then expand the current state into a number of future states by determining all the possible actions that can follow from that state, then expand all the future states into their own future states, and so on. When the AI cannot expand the states any further (stops the expansion), it has created a state space consisting of everything that could happen in the future. AI can take advantage of a state space not only to rank all possibilities, predicting which future states are morel likely than others, but also for exploring decisions it can make to reach its goal in the best way. This is known as the state-space search. Working with a state space requires use of both particular data structures, such as trees and graphs, as well as algorithms, with the two favored versions used to efficiently explore graphs being breadth-first search and deep-first search. A tree is formed by starting with a "root node," and each added item is itself a node, which connects to one or two subsequent nodes using links. Nodes that support other nodes are branch nodes. The final ending points on a tree are leaf nodes. In the relationship between nodes, the node that does the supporting is the parent, and the node that is supported is the child. On the other hand, when dealing with graphs, nodes can have more than two (binary) connections, so that graphs are a kind of "tree extension." Moreover, the nodes in a graph can connect in any direction—not just from parent to child.
The State of AI in Business The topic of AI and its potential to revolutionise business growth and performance is frequently discussed, but what is the overall outlook for its practical use in the business world? 44% of private sector companies plan to invest in AI systems in 2023 35% of companies are using AI and 42% of companies are exploring AI for its implementation in the future 91.5% of leading businesses invest in AI on an ongoing basis Customer satisfaction is expected to grow by 25% by 2023 in organisation that use AI
Playing Adversarial Games In 1987, Ronald Rivest introduced an approach called the min-max approximation (which is supposedly superior to minimax search with "alpha-beta pruning," which is a smart way to propagate values up the tree hierarchy in complex state spaces, thereby limiting computaions). According to Mueller and Massaron, since then, this algorithm and its variants have powered many competitive games, along with recent game-playing advances, such as AlphaGo from Google DeepMind, which uses an approach that echoes the min-max approximation, which is also found in the Wargames film of 1983. However, not all games feature compact state-space trees. Accordingly, when branches are in the numbers of millions, it is necessary to prune them and shorten calculations. In the end, no machine, no matter how powerful, can enumerate all the possibilities that spring from real-life situations (as opposed to game situations, which have fixed rules and are therefore more predictable). Such pruning and shortening can be accomplished using local search and heuristics. (More on this in the future.)
(Semafor) Brain-inspired chip boosts AI speed A chip inspired by the human brain is 20 times as fast as any existing one at running artificial-intelligence programs. IBM’s NorthPole unites processor and memory within one chip, in the same way that human neurons both compute and store data, IEEE Spectrum reported. That combination is part of why brains are vastly more energy-efficient than computers. NorthPole, similarly, is 25 times more efficient than standard chips when used to run neural networks, which are also inspired by the architecture of the human brain. IBM says the new chip will have important uses in autonomous vehicles, image recognition, and robotics.
Think work meetings are bad now, wait until AI gets involved. Are AI Chatbots About to Make Work Meetings Worse Than Ever? Big Tech companies are launching new tools to make work calls easier—but with privacy issues and rogue bots, they could end up causing more harm than good. https://www.thedailybeast.com/will-microsofts-copilot-ai-make-work-meetings-worse-than-ever
(Semafor) April 16th, 2024 New energy-efficient AI chip TSINGHUA UNIVERSITY IN BEIJING/BEIJING NATIONAL RESEARCH CENTER FOR INFORMATION SCIENCE AND TECHNOLOGY A microchip that uses light instead of electricity could cut energy demand for artificial intelligence systems by more than 99%. “Optical computing” that uses these photonic chips has been proposed as a possible solution to AI’s growing energy demand, and to boost the speed at which chips can operate by passing messages at the speed of light. The new chip, Taichi, produced by Chinese researchers, runs AI tasks as well as silicon chips, using a thousandth as much energy. “Optical neural networks are no longer toy models,” an engineering professor told IEEE Spectrum. “They can now be applied in real-world tasks.”
(Semafor) 4/22/24 Computer mimics human brain activity Courtesy Intel Intel unveiled the largest ever computer designed to mimic human brain activity. “Neuromorphic” computers use artificial neurons that both store data and compute information, as neurons in our brain seem to, unlike traditional computers in which memory and processing are separate. Removing the need to shuttle data between the two components should avoid bottlenecks and reduce energy consumption: Intel claims that “Hala Point” will need 100 times less energy to an equivalently powerful conventional machine. It may also provide a path to more general artificial intelligence: “The dream is that one day neuromorphic computing will enable us to make brain-like models,” one engineer told New Scientist.
Can Artificial Intelligence Language Models Bridge the Earth System Knowledge Gap? Artificial intelligence (AI) language models present a promising solution to make Earth system science user-friendly for experts and non-experts, a critical step towards effective climate action. WMO Secretary-General Professor Celeste Saulo has highlighted the urgent need to strengthen communication efforts and to ensure universal access to information for preparedness when facing extreme weather, climate and water events, especially for the most vulnerable. This will require making Earth system science user-friendly for experts and non-experts, a critical step towards effective climate action and ensuring that meteorological, climatological and hydrological services receive the political backing they deserve. Artificial intelligence (AI) language models present a promising solution to these challenges. The WMO houses a comprehensive library of reports of immense scientific value. However, the information an individual, businessperson or policy maker may need to comprehend the magnitude of the climate crisis might be scattered across various digital platforms or be in language that is too challenging to comprehend and act upon. ChatClimate, a specialized AI akin to ChatGPT focusing on WMO and IPCC reports, offers a solution by providing instant access to Earth system science information, moving away from the cumbersome process of navigating extensive PDF documents. ChatClimate’s mission is to make climate risks understandable and climate information more accessible to a wider audience. In today’s digital era, distinguishing between accurate and misleading climate information is essential, given its profound impact on public perception and policy decisions. Natural language processing and automated fact-checking aims to verify claims using reliable factual sources. The Climinator, an acronym for {CLI}mate {M}ediator for {IN}formed {A}nalysis and {T}ransparent {O}bjective {R}easoning represents such an innovation. It synthesizes various scientific viewpoints to produce robust, evidence-based evaluations. The model shows remarkable accuracy in testing claims and demonstrates its capability to foster a consensus even when integrating perspectives from climate science skeptics. While the research has its limitations and demands cautious interpretation, it underscores the potential of AI in reconciling diverse viewpoints into factual conclusions. AI language models, such as ChatClimate and the Climinator, offer innovative solutions for the dissemination of Earth system science and for combating misinformation. By making scientific information more accessible and validating claims against scientific consensus, these tools can potentially advance climate action efforts for National Meteorological and Hydrological Services (NMHSs) around the world. The commitment to integrating AI into Earth system science reflects a forward-thinking approach to leverage technology in addressing some of the most pressing global challenge.
https://techcrunch.com/2024/05/20/m...utm_medium=newsletter&utm_source=morning_brew Microsoft wants to make Windows an AI operating system, launches Copilot+ PCs Kyle Wiggers 4:52 PM PDT • May 20, 2024